Pub Date : 2022-10-28DOI: 10.1109/ICUS55513.2022.9987243
Zhiwen Huang, Hongxu Li, Yuanfu Zhong, Qian Su, Na Lv, Yingchao Zhang
Historical war experience demonstrates that crashing critical supply facility has become an efficient tactic during wartime, so it is of great strategic significance to design an efficient and reliable emergency logistics network in face of the uncertainty at war. This paper presents a scenario-based modelling method by considering both demand uncertainty and facility failure scenarios in the mathematical model and the immune genetic algorithm (IGA) is applied to solve it. A set of numerical experiments has been conducted to verify the model and effectiveness of IGA. Finally, the sensitivity analysis results show facility failure probability has a greater impact on the final location decision, which provide model and approach for the location-allocation problem during wartime.
{"title":"Reliability Facility Location with Fuzzy Demand and Failure Scenarios","authors":"Zhiwen Huang, Hongxu Li, Yuanfu Zhong, Qian Su, Na Lv, Yingchao Zhang","doi":"10.1109/ICUS55513.2022.9987243","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9987243","url":null,"abstract":"Historical war experience demonstrates that crashing critical supply facility has become an efficient tactic during wartime, so it is of great strategic significance to design an efficient and reliable emergency logistics network in face of the uncertainty at war. This paper presents a scenario-based modelling method by considering both demand uncertainty and facility failure scenarios in the mathematical model and the immune genetic algorithm (IGA) is applied to solve it. A set of numerical experiments has been conducted to verify the model and effectiveness of IGA. Finally, the sensitivity analysis results show facility failure probability has a greater impact on the final location decision, which provide model and approach for the location-allocation problem during wartime.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133822615","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-28DOI: 10.1109/ICUS55513.2022.9987061
Zhiyu Duan, Xiaolin Chen, Ziyuan Xie
This paper solves the distributed task allocation problem based on the danger-broadcast algorithm for unmanned vehicles with the Consensus-Based Bundle Algorithm (CBBA). By defining the dangerous region that is found by the flying unmanned vehicle, the traditional Consensus-Based Bundle Algorithm is extended with the danger broadcast, which is named as Danger-broadcast Consensus-Based Bundle Algorithm (DBCBBA). In the unknown task areas, the time discount and distance are used to describe the task's revenue function. Compared with the traditional CBBA, the proposed algorithm can be adopted to a real and complex environment, and more unmanned vehicles are saved by distributed danger-broadcast architecture. In the end, simulation experiments demonstrate the performance of the proposed algorithm for the unknown danger zone.
{"title":"Consensus-Based Bundle Algorithm-Based Task Allocation for Unmanned Vehicles in Dangerous Environment","authors":"Zhiyu Duan, Xiaolin Chen, Ziyuan Xie","doi":"10.1109/ICUS55513.2022.9987061","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9987061","url":null,"abstract":"This paper solves the distributed task allocation problem based on the danger-broadcast algorithm for unmanned vehicles with the Consensus-Based Bundle Algorithm (CBBA). By defining the dangerous region that is found by the flying unmanned vehicle, the traditional Consensus-Based Bundle Algorithm is extended with the danger broadcast, which is named as Danger-broadcast Consensus-Based Bundle Algorithm (DBCBBA). In the unknown task areas, the time discount and distance are used to describe the task's revenue function. Compared with the traditional CBBA, the proposed algorithm can be adopted to a real and complex environment, and more unmanned vehicles are saved by distributed danger-broadcast architecture. In the end, simulation experiments demonstrate the performance of the proposed algorithm for the unknown danger zone.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"78 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123994513","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-28DOI: 10.1109/ICUS55513.2022.9987228
Longfei Yue, Rennong Yang, Jialiang Zuo, Y. Zhang, Xiaoru Zhao, Mengda Yan
Aircraft maneuver recognition is a key issue in unmanned aerial vehicle (UAV) intelligent air combat. Aiming at inefficiency of high-dimensional time-series maneuver data analysis and low recognition accuracy of traditional methods, an end-to-end regularized autoencoder-support vector machine (RAE-SVM) method is proposed. This method combines powerful feature extraction capability of unsupervised learning based autoencoder with superior classification performance of supervised learning based SVM. According to the change rule of maneuver data and prior expert knowledge, the maneuver recognition dataset based on time period feature data is constructed. The generalization performance of RAE network and the accuracy of the model are improved by introducing regularization. The aircraft maneuver recognition model based on RAE-SVM is constructed and verified by the maneuver recognition dataset. The simulation results show that the accuracy of model recognition is as high as 92.75%. The trained model only takes 2 milliseconds to recognize a set of maneuver data and meets the near real-time requirements. Therefore, the proposed approach in this work can quickly and accurately recognize aircraft maneuver without relying on expert experience, which has certain practical value.
{"title":"Aircraft Maneuver Intelligent Recognition: An End-to-end Regularized Autoencoder and SVM Approach","authors":"Longfei Yue, Rennong Yang, Jialiang Zuo, Y. Zhang, Xiaoru Zhao, Mengda Yan","doi":"10.1109/ICUS55513.2022.9987228","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9987228","url":null,"abstract":"Aircraft maneuver recognition is a key issue in unmanned aerial vehicle (UAV) intelligent air combat. Aiming at inefficiency of high-dimensional time-series maneuver data analysis and low recognition accuracy of traditional methods, an end-to-end regularized autoencoder-support vector machine (RAE-SVM) method is proposed. This method combines powerful feature extraction capability of unsupervised learning based autoencoder with superior classification performance of supervised learning based SVM. According to the change rule of maneuver data and prior expert knowledge, the maneuver recognition dataset based on time period feature data is constructed. The generalization performance of RAE network and the accuracy of the model are improved by introducing regularization. The aircraft maneuver recognition model based on RAE-SVM is constructed and verified by the maneuver recognition dataset. The simulation results show that the accuracy of model recognition is as high as 92.75%. The trained model only takes 2 milliseconds to recognize a set of maneuver data and meets the near real-time requirements. Therefore, the proposed approach in this work can quickly and accurately recognize aircraft maneuver without relying on expert experience, which has certain practical value.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"111 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124497262","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-28DOI: 10.1109/ICUS55513.2022.9987012
Tao Xu, Xiaojian Yi, G. Wen, Z. Duan
This paper studies the formation control problem of networked second-order integrator systems from a distributed event-driven perspective, where the unknown nonlinearity is involved in the system model. An event-driven formation control scheme, which consists of a fully distributed event-driven control protocol and a fully distributed triggering function, is developed in this paper. In order to handle the nonlinear dynamics, the approximation property of the neural-network control is utilized. Adaptive gains instead of constant gains are designed in the control protocol, making the control scheme fully distributed. Using the proposed triggering mechanism, the control torque of each agent is a piecewise constant function, which is updated discontinuously and asynchronously. Moreover, the information topologies among different agents are directed and the prescribed formation configuration is time-varying. These settings are more practical, but brings some difficulties to control scheme design and theoretical analysis. It is shown that under the developed control scheme, each agent can achieve the prescribed formation configuration without causing the undesired Zeno behavior. Finally, numerical simulation is performed to confirm the validity of the main theorems.
{"title":"Fully Distributed Event-Driven Formation Control Over Directed Information Topologies","authors":"Tao Xu, Xiaojian Yi, G. Wen, Z. Duan","doi":"10.1109/ICUS55513.2022.9987012","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9987012","url":null,"abstract":"This paper studies the formation control problem of networked second-order integrator systems from a distributed event-driven perspective, where the unknown nonlinearity is involved in the system model. An event-driven formation control scheme, which consists of a fully distributed event-driven control protocol and a fully distributed triggering function, is developed in this paper. In order to handle the nonlinear dynamics, the approximation property of the neural-network control is utilized. Adaptive gains instead of constant gains are designed in the control protocol, making the control scheme fully distributed. Using the proposed triggering mechanism, the control torque of each agent is a piecewise constant function, which is updated discontinuously and asynchronously. Moreover, the information topologies among different agents are directed and the prescribed formation configuration is time-varying. These settings are more practical, but brings some difficulties to control scheme design and theoretical analysis. It is shown that under the developed control scheme, each agent can achieve the prescribed formation configuration without causing the undesired Zeno behavior. Finally, numerical simulation is performed to confirm the validity of the main theorems.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116684304","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Generative adversarial networks (GANs) are well-known for their powerful generation ability. In recent years, GANs have been applied in the field of aerodynamic shape optimization (ASO). However, the existing airfoil generation methods based on GANs can only generate a discrete sequence of coordinates corresponding to fixed abscissas, and cannot be applied to the scenarios that generate airfoils directly. In this paper, class function / shape function transformation (CST), a parameterization method of the airfoil that forms a good representation of the geometric shape of the airfoil, is combined with GANs. Therefore, a CST-GANs method is proposed that can directly generate the CST parameterized variables of the airfoil instead of a sequence of airfoil points. Given the abscissa and parameterized variables, the corresponding coordinate can be calculated by CST expression. On the other hand, CST-GANs can generate airfoil geometry with smooth surface without intro-ducing the Bézier curve or the Savitzky-Golay filter. Experiments show that CST-GANs is a promising model, which can not only generate smoother airfoils with fewer neural network parameters but also generate more diverse airfoils.
生成对抗网络(GANs)以其强大的生成能力而闻名。近年来，gan已被应用于气动形状优化(ASO)领域。然而，现有的基于gan的翼型生成方法只能生成与固定横坐标对应的离散坐标序列，不能应用于直接生成翼型的场景。本文将类函数/形状函数变换(class function / shape function transformation, CST)这一能够很好地表征翼型几何形状的参数化方法与gan相结合。因此，提出了一种CST- gans方法，该方法可以直接生成翼型的CST参数化变量，而不是翼型点序列。给定横坐标和参数化变量，可以通过CST表达式计算相应的坐标。另一方面，cst - gan可以产生表面光滑的翼型几何形状，而无需引入bsamizier曲线或Savitzky-Golay滤波器。实验表明，cst - gan模型不仅可以用更少的神经网络参数生成更光滑的翼型，而且可以生成更多样化的翼型，是一种很有前途的模型。
{"title":"CST-GANs: A Generative Adversarial Network Based on CST Parameterization for the Generation of Smooth Airfoils","authors":"Jinxing Lin, Chenliang Zhang, Xiaoye Xie, Xingyu Shi, Xiaoyu Xu, Yanhui Duan","doi":"10.1109/ICUS55513.2022.9987080","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9987080","url":null,"abstract":"Generative adversarial networks (GANs) are well-known for their powerful generation ability. In recent years, GANs have been applied in the field of aerodynamic shape optimization (ASO). However, the existing airfoil generation methods based on GANs can only generate a discrete sequence of coordinates corresponding to fixed abscissas, and cannot be applied to the scenarios that generate airfoils directly. In this paper, class function / shape function transformation (CST), a parameterization method of the airfoil that forms a good representation of the geometric shape of the airfoil, is combined with GANs. Therefore, a CST-GANs method is proposed that can directly generate the CST parameterized variables of the airfoil instead of a sequence of airfoil points. Given the abscissa and parameterized variables, the corresponding coordinate can be calculated by CST expression. On the other hand, CST-GANs can generate airfoil geometry with smooth surface without intro-ducing the Bézier curve or the Savitzky-Golay filter. Experiments show that CST-GANs is a promising model, which can not only generate smoother airfoils with fewer neural network parameters but also generate more diverse airfoils.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116692755","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-28DOI: 10.1109/ICUS55513.2022.9986530
Dewei Deng, Ziqian Xiong, Chuan Wang, Hao Liu
Track similarity analysis is an important part of historical track big data mining analysis, which provides a basis for track clustering, target recognition, and target activity law analysis. The paper adopts the improved longest common subsequence method to measure the similarity between different tracks and complete track matching. Firstly, we establish a historical track database and establish a raster spatial index for the key sea areas of concern. Secondly, we preprocess the track data and unify the data of different sampling scales into one space-time dimension through sparse or interpolation, retrieve historical tracks that are geographically close to the current track, and use the improved longest common subsequence method to calculate the matching similarity probability between the current track and the historical track. Finally, the track clustering and activity rule mining analysis are completed according to the calculated maximum matching probability. The simulation experiment verifies that the algorithm can effectively calculate and complete the correlation matching of target tracks, which provides an effective basis for target track clustering and activity law mining and analysis.
{"title":"Track Matching Method of Sea Surface Targets Based on Improved Longest Common Subsequence Algorithm","authors":"Dewei Deng, Ziqian Xiong, Chuan Wang, Hao Liu","doi":"10.1109/ICUS55513.2022.9986530","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9986530","url":null,"abstract":"Track similarity analysis is an important part of historical track big data mining analysis, which provides a basis for track clustering, target recognition, and target activity law analysis. The paper adopts the improved longest common subsequence method to measure the similarity between different tracks and complete track matching. Firstly, we establish a historical track database and establish a raster spatial index for the key sea areas of concern. Secondly, we preprocess the track data and unify the data of different sampling scales into one space-time dimension through sparse or interpolation, retrieve historical tracks that are geographically close to the current track, and use the improved longest common subsequence method to calculate the matching similarity probability between the current track and the historical track. Finally, the track clustering and activity rule mining analysis are completed according to the calculated maximum matching probability. The simulation experiment verifies that the algorithm can effectively calculate and complete the correlation matching of target tracks, which provides an effective basis for target track clustering and activity law mining and analysis.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"24 12","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114116412","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-28DOI: 10.1109/ICUS55513.2022.9986647
Wei Fu, Xingfa Shen, Wenhui Zhou, A. Zhdanov, Chuntong Geng
Light field imaging is an important achievement in visual information exploration in recent years, which can capture more abundant visual information from the real world. However, most existing light field image quality assessment (LF-IQA) indicators rely heavily on feature extraction based on high complexity statistics in quality evaluation tasks, which is not comprehensive in future modern applications or power-limited devices. To solve this problem, the research puts forward a light field image quality evaluation method based on stereo vision. By introducing a brand-new light field image coding method and training the neural network, the purpose of image quality evaluation is finally achieved. Some experiments show that our model achieves good results in open source data sets.
{"title":"A Light Field Image Quality Assessment Method Based on Stereo Vision","authors":"Wei Fu, Xingfa Shen, Wenhui Zhou, A. Zhdanov, Chuntong Geng","doi":"10.1109/ICUS55513.2022.9986647","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9986647","url":null,"abstract":"Light field imaging is an important achievement in visual information exploration in recent years, which can capture more abundant visual information from the real world. However, most existing light field image quality assessment (LF-IQA) indicators rely heavily on feature extraction based on high complexity statistics in quality evaluation tasks, which is not comprehensive in future modern applications or power-limited devices. To solve this problem, the research puts forward a light field image quality evaluation method based on stereo vision. By introducing a brand-new light field image coding method and training the neural network, the purpose of image quality evaluation is finally achieved. Some experiments show that our model achieves good results in open source data sets.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115179913","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-28DOI: 10.1109/ICUS55513.2022.9986540
Linfeng Su, Jinbo Wang, Zhenwei Ma, Hongbo Chen
Powered landing of reusable rocket is an advanced technology to achieve pinpoint landing (the norm of position error < 5 m and velocity error < 2 m/s) while satisfying a series of highly nonlinear constraints. A major challenge is guaranteeing fuel-optimal and convergence when solving rocket powered landing problem. In this manuscript, a real-time feedback guidance algorithm based on deep reinforcement learning is developed. The proposed method maps state directly to thrust control commands. The first contribution of this paper is to use multi-stage reward function to eliminate the negative effects triggered by design guidance law, thereby significantly enhancing fuel-optimal performance. Another contribution is that a model pre-training framework based on imitation learning is presented to improve model convergence by fitting optimal data. Numerical experiments show that the nearly fuel-optimal trajectories generated by the proposed algorithm successfully achieve pinpoint landing from random initial states.
{"title":"Real-time Guidance for Powered Landing of Reusable Rockets via Deep Reinforcement Learning","authors":"Linfeng Su, Jinbo Wang, Zhenwei Ma, Hongbo Chen","doi":"10.1109/ICUS55513.2022.9986540","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9986540","url":null,"abstract":"Powered landing of reusable rocket is an advanced technology to achieve pinpoint landing (the norm of position error < 5 m and velocity error < 2 m/s) while satisfying a series of highly nonlinear constraints. A major challenge is guaranteeing fuel-optimal and convergence when solving rocket powered landing problem. In this manuscript, a real-time feedback guidance algorithm based on deep reinforcement learning is developed. The proposed method maps state directly to thrust control commands. The first contribution of this paper is to use multi-stage reward function to eliminate the negative effects triggered by design guidance law, thereby significantly enhancing fuel-optimal performance. Another contribution is that a model pre-training framework based on imitation learning is presented to improve model convergence by fitting optimal data. Numerical experiments show that the nearly fuel-optimal trajectories generated by the proposed algorithm successfully achieve pinpoint landing from random initial states.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125594619","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In order to improve the accuracy of object detection based on deep learning for military vehicles., an object detection algorithm based on image style transfer and domain adversarial learning is proposed. Aiming at the problem of the small number of military vehicle images., a military vehicle image dataset is constructed by using games., miniature models and spider technology. However., due to the visual difference between these images., the accuracy of the trained detection model is poor when it is directly used to detect the actual collected images. CycleGAN is used for image style transfer to obtain images with a similar style to military vehicle images at the data level. Domain adversarial learning is used to optimize the one-stage object detection algorithm to make the network learn domain invariant features and reduce the domain discrepancy at the feature level. The proposed algorithm is implemented with YOLOv5s as an example. The test results show that the proposed algorithm improves the average precision (AP) by 5.7% without increasing the amount of inference computation., compared with the YOLOv5s algorithm.
{"title":"An Object Detection Algorithm for Military Vehicles Based on Image Style Transfer and Domain Adversarial Learning","authors":"Yubeibei Zhou, Jiulu Gong, Weijian Lu, Naiwei Gu, Kuiqi Chong, Zepeng Wang","doi":"10.1109/ICUS55513.2022.9987190","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9987190","url":null,"abstract":"In order to improve the accuracy of object detection based on deep learning for military vehicles., an object detection algorithm based on image style transfer and domain adversarial learning is proposed. Aiming at the problem of the small number of military vehicle images., a military vehicle image dataset is constructed by using games., miniature models and spider technology. However., due to the visual difference between these images., the accuracy of the trained detection model is poor when it is directly used to detect the actual collected images. CycleGAN is used for image style transfer to obtain images with a similar style to military vehicle images at the data level. Domain adversarial learning is used to optimize the one-stage object detection algorithm to make the network learn domain invariant features and reduce the domain discrepancy at the feature level. The proposed algorithm is implemented with YOLOv5s as an example. The test results show that the proposed algorithm improves the average precision (AP) by 5.7% without increasing the amount of inference computation., compared with the YOLOv5s algorithm.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126182817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-28DOI: 10.1109/ICUS55513.2022.9986982
Jiarui Ma, Jinbo Wang, Qiliang Zhang
This paper introduces the extrapolated proportional integral projected gradient (ePIPG) method, a newly developed first-order method, for second-order cone programming problems (SOCP), and its customized implementation to solve the 3-DoF powered-descent guidence problem. The ePIPG solvers parse the problem parametets automatically to fully utilize the problem sparse structure. In addition, different from the classic interior-point method, being a first-order method, ePIPG only relies on simple algebra operations, for example, projection onto convex sets, multiplication of matrix and vector and vectors addition. The numerical experiment shows that the customized ePIPG solver is significantly faster than ECOS and GUROBI, two advanced convex optimization solvers, thus is a potential method for future rocket landing missions.
{"title":"Customized First Order Convex Optimization Algorithm for Powered-Descent Guidance","authors":"Jiarui Ma, Jinbo Wang, Qiliang Zhang","doi":"10.1109/ICUS55513.2022.9986982","DOIUrl":"https://doi.org/10.1109/ICUS55513.2022.9986982","url":null,"abstract":"This paper introduces the extrapolated proportional integral projected gradient (ePIPG) method, a newly developed first-order method, for second-order cone programming problems (SOCP), and its customized implementation to solve the 3-DoF powered-descent guidence problem. The ePIPG solvers parse the problem parametets automatically to fully utilize the problem sparse structure. In addition, different from the classic interior-point method, being a first-order method, ePIPG only relies on simple algebra operations, for example, projection onto convex sets, multiplication of matrix and vector and vectors addition. The numerical experiment shows that the customized ePIPG solver is significantly faster than ECOS and GUROBI, two advanced convex optimization solvers, thus is a potential method for future rocket landing missions.","PeriodicalId":345773,"journal":{"name":"2022 IEEE International Conference on Unmanned Systems (ICUS)","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126244750","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：